A Universal Music Translation Network (Facebook AI Research, 2018) is a deep-learning model for music style transfer. It uses a shared encoder $E$ together with per-domain WaveNet decoders $D_j$ to transfer timbre and playing style (e.g., between jazz, traditional, and other genres) while preserving the underlying musical content. Training combines a classification loss, which pushes the encoder toward a domain-confused (style-invariant) representation via a domain classifier $C$, with a reconstruction loss that keeps each decoder's output faithful to the original input. This combination showed that musical style can be disentangled from content, and the model is widely cited as an influential baseline for music style transfer.